4 Sessions

4 Sessions4 Sessions
Marian HackMan MarinovMarian HackMan Marinov
OpenFestOpenFest

1st1st
Increasing the performance usingIncreasing the performance using
SSE, AVX* and FMA extensionsSSE, AVX* and FMA extensions
https://jobs.siteground.bg/#devopshttps://jobs.siteground.bg/#devops

● AVX - Advanced Vector Extensions
● AVX2 - 256bit integers
– FMA - Fused multiply-accumulate
● AVX-512 - 512bit integers
● SSE - Streaming SIMD Extensions
SIMD - Single Instruction Multiple Data

● Exploit AVX for matrix multiplication
● Exploit SSE
- for binary operations on multiple inputs
- for populating multiple registers with single
instructions
● AVX-512 for prefetching data

Why does it work?Why does it work?
Vectorization Vectorization
#define MAX 1000000#define MAX 1000000
int a[256], b[256], c[256];int a[256], b[256], c[256];
int main () {int main () {
int i,j;int i,j;
for (j=0; j<MAX; j++){for (j=0; j<MAX; j++){
for (i=0; i<256; i++){for (i=0; i<256; i++){
a[i] = b[i] + c[i];a[i] = b[i] + c[i];
}}
return 0;return 0;

A[1]A[1] not usednot used not usednot used not usednot used
B[1]B[1] not usednot used not usednot used not usednot used
+
C[1]C[1] not usednot used not usednot used not usednot used
3x 32-bit unused integers

A[3]A[3]
B[3]B[3]
+
C[3]C[3]
A[2]A[2] A[1]A[1] A[0]A[0]
B[2]B[2] B[1]B[1] B[0]B[0]
C[2]C[2] C[1]C[1] C[0]C[0]

$ gcc -fopt-info-vec sort.c –O2 –ftree-vectorize$ gcc -fopt-info-vec sort.c –O2 –ftree-vectorize
$ gcc -fopt-info-vec sort.c –O3$ gcc -fopt-info-vec sort.c –O3
https://github.com/VictorRodriguez/autofdo_tutorial/blob/master/sort.c
0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0
with vectorization without vectorization (O3)
1.0x
15.9x

AVX512AVX512
$ gcc -O3 sanity.c -fopt-info-vec -mavx2 -o sanity$ gcc -O3 sanity.c -fopt-info-vec -mavx2 -o sanity
511 -> 256 255 -> 128 127 -> 0
Intel AVX
512
Intel AVX2/
Intel AVX
SSE
XMM0 YMM0 ZMM0
XMM1 YMM1 ZMM1
XMM2 YMM2 ZMM2
XMM3 YMM3 ZMM3
XMM4 YMM4 ZMM4
XMM5 YMM5 ZMM5
XMM6 YMM6 ZMM6

$ gcc -O3 sanity.c -fopt-info-vec -mavx2 -o sanity$ gcc -O3 sanity.c -fopt-info-vec -mavx2 -o sanity
0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 18.0 20.0 22.0
with vectorization without vectorization (O3)
1.0x
23.2x
15.9x
AVX2

It's complicatedIt's complicated

Intel Clear LinuxIntel Clear Linux
https://github.com/clearlinux/makefmvpatchhttps://github.com/clearlinux/makefmvpatch
https://github.com/clearlinuxpkgshttps://github.com/clearlinuxpkgs
https://clearlinux.org/https://clearlinux.org/
* Modified glibc* Modified glibc
* Modified Python package* Modified Python package
* Modified R package* Modified R package

2nd2nd
BPF BCC toolsBPF BCC tools
for performance analysisfor performance analysis

What is BPF?
# tcpdump host 127.0.0.1 and port 22 -d
(000) ldh [12] Optimizes packet filter
(001) jeq #0x800 jt 2 jf 18 performance
(002) ld [26]
(003) jeq #0x7f000001 jt 6 jf 4
(004) ld [30] 2 x 32-bit registers
(005) jeq #0x7f000001 jt 6 jf 18 & scratch memory
(006) ldb [23]
(007) jeq #0x84 jt 10 jf 8
(008) jeq #0x6 jt 10 jf 9
(009) jeq #0x11 jt 10 jf 18 User-defined bytecode
(010) ldh [20] executed by an in-kernel
(011) jset #0x1fff jt 18 jf 12 sandboxed virtual machine
(012) ldxb 4*([14]&0xf)
(013) ldh [x + 14][...] Steven McCanne and Van Jacobson, 1993

What is eBPF?
/* Register numbers */
enum {
BPF_REG_0 = 0,
BPF_REG_1,
BPF_REG_2,
BPF_REG_3, 10 x 64-bit registers
BPF_REG_4, maps (hashes)
BPF_REG_5, actions
BPF_REG_6,
BPF_REG_7,
BPF_REG_8,
BPF_REG_9,
BPF_REG_10,
__MAX_BPF_REG,
};

What is eBPF?
struct bpf_insn prog[] = {
BPF_MOV64(BPF_REG_6, BPF_REG_1),
BPF_LD_ABS(BPF_B, ETH_HLEN + offsetof(struct iphdr, protocol), /* R0 = ip-
>proto */
BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG0, -4), /* *(u32 *)(fp - 4) =
R0 */
BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), /* R2 = fp - 4 */
BPF_LD_MAP_FDD(BPF_REG_1, map_fd),
BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0,
BPF_FUNC_map_lookup_elem),
BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2),
BPF_MOV64_IMM(BPF_REG_1, 1), /* R1 = 1 */
BPF_RAW_INSN(BPF_STX | BPF_XADD | BPF_DW, BPF_REG_0,
BPF_REG_1, 0, 0), /* xadd R0 += R1 */
BPF_MOV64_IMM(BPF_REG_0, 0), /* R0 = 0 */
BPF_EXIT_INSN(),
};

How does it work?

What else can you do with it?

Where are these tools?
https://github.com/iovisor/bcc
Brendan Gregg
Senior Performance Architect, Netflix

Some examples
# ./execsnoop
PCOMM PID RET ARGS
supervise 9660 0 ./run
supervise 9661 0 ./run
mkdir 9662 0 /bin/mkdir -p ./main
run 9663 0 ./run
[...]

Some examples
# ./opensnoop
PID COMM FD ERR PATH
1565 redis-server 5 0 /proc/1565/stat
1603 snmpd 9 0 /proc/net/dev
1603 snmpd 11 0 /proc/net/if_inet6
1603 snmpd -1 2 /sys/class/net/eth0/device/vendor
1603 snmpd 11 0
/proc/sys/net/ipv4/neigh/eth0/retrans_time_ms
1603 snmpd 11 0
/proc/sys/net/ipv6/neigh/eth0/retrans_time_ms
1603 snmpd 11 0
/proc/sys/net/ipv6/conf/eth0/forwarding
[...]

Some examples
# ./cachestat
HITS MISSES DIRTIES READ WRITE BUFFERS CACHED
HIT% HIT% MB
1074 44 13 94.9% 2.9% 1 223
2195 170 8 92.5% 6.8% 1 143
182 53 56 53.6% 1.3% 1 143
62480 40960 20480 40.6% 19.8% 1 223
7 2 5 22.2% 22.2% 1 223
348 0 0 100.0% 0.0% 1 223
[...]

Some examples
# ./biolatency
Tracing block device I/O... Hit Ctrl-C to end.
^C
usecs : count distribution
0 -> 1 : 0 | |
2 -> 3 : 0 | |
4 -> 7 : 0 | |
8 -> 15 : 0 | |
16 -> 31 : 0 | |
32 -> 63 : 0 | |
64 -> 127 : 1 | |
128 -> 255 : 12 |******** |
256 -> 511 : 15 |********** |
512 -> 1023 : 43 |******************************* |
1024 -> 2047 : 52 |**************************************|
2048 -> 4095 : 47 |********************************** |
4096 -> 8191 : 52 |**************************************|
8192 -> 16383 : 36 |************************** |
16384 -> 32767 : 15 |********** |
32768 -> 65535 : 2 |* |
65536 -> 131071 : 2 |* |

Some examples
# ./biosnoop
TIME(s) COMM PID DISK T SECTOR BYTES LAT(ms)
0.000004001 supervise 1950 xvda1 W 13092560 4096 0.74
0.000178002 supervise 1950 xvda1 W 13092432 4096 0.61
0.001469001 supervise 1956 xvda1 W 13092440 4096 1.24
0.001588002 supervise 1956 xvda1 W 13115128 4096 1.09
1.022346001 supervise 1950 xvda1 W 13115272 4096 0.98
1.022568002 supervise 1950 xvda1 W 13188496 4096 0.93
[...]

Some examples
# ./runqlat
Tracing run queue latency... Hit Ctrl-C to end.
usecs : count distribution
0 -> 1 : 233 |*********** |
2 -> 3 : 742 |************************************ |
4 -> 7 : 203 |********** |
8 -> 15 : 173 |******** |
16 -> 31 : 24 |* |
32 -> 63 : 0 | |
64 -> 127 : 30 |* |
128 -> 255 : 6 | |
256 -> 511 : 3 | |
512 -> 1023 : 5 | |
1024 -> 2047 : 27 |* |
2048 -> 4095 : 30 |* |
4096 -> 8191 : 20 | |
8192 -> 16383 : 29 |* |
16384 -> 32767 : 809 |****************************************|
32768 -> 65535 : 64 |***

3rd3rd
Insecurity of today's Insecurity of today's
computers. computers.
Ring 2 firmware and UEFI, Ring 2 firmware and UEFI,
and why we wouldn't want and why we wouldn't want
themthem

Linuxcon 2017 NERFLinuxcon 2017 NERF

4th4th
Comparison between the Comparison between the
functionality of the best functionality of the best
known Nginx distributionsknown Nginx distributions
NginxNginx, , OpenRestyOpenResty and and
TengineTengine

Nginx is one of the fastestNginx is one of the fastest
web servers in the worldweb servers in the world

How to get it?How to get it?
Distribution package Distribution package
other repos with prebuild other repos with prebuild
packagespackages

How to get it?How to get it?
Manual compilation Manual compilation
go with Nginx plus go with Nginx plus

Alternatives?Alternatives?
OpenResty OpenResty
Tengine Tengine

OpenRestyOpenResty
   OpenResty® is a dynamic web platform OpenResty® is a dynamic web platform
based on NGINX and LuaJIT.based on NGINX and LuaJIT.
   a good source for high quality Nginx a good source for high quality Nginx
modulesmodules
   25 different nginx modules 25 different nginx modules
  https://openresty.org/en/https://openresty.org/en/
  https://github.com/openresty/https://github.com/openresty/

OpenRestyOpenResty
* highlights:* highlights:
      sregexsregex
      headersmoreheadersmore
           clear headers on input clear headers on input
           clear or replace headers on output clear or replace headers on output
      replacefilterreplacefilter
           regexp replace BODY filter regexp replace BODY filter

OpenRestyOpenResty
I believe that it is the best I believe that it is the best
web application platform web application platform
you can directly useyou can directly use

TengineTengine
this is the web server that this is the web server that
Alibaba runs onAlibaba runs on
its main purpose is its main purpose is
performanceperformance
its a collection of different its a collection of different
nginx modulesnginx modules

TengineTengine
  * Proxy/Load balancing* Proxy/Load balancing
       Dynamic Upstream updates Dynamic Upstream updates
       Upstream domain resolver Upstream domain resolver
       Limit upstream tries Limit upstream tries
       Upstream check module Upstream check module
       Upstream keepalive timeout Upstream keepalive timeout
       Consistent hash module Consistent hash module
       Session sticky module Session sticky module
       Slice module Slice module

TengineTengine
  * Filters* Filters
       Concat Concat
       Headers Headers
       Footer Footer
       Trim Trim
       Reqstat Reqstat
       TFS TFS
       User agent User agent

ConclusionConclusion
if you are preparing a load if you are preparing a load
balancer/proxy, go with Tenginebalancer/proxy, go with Tengine
it you are preparing a web it you are preparing a web
application server, go with application server, go with
OpenRestyOpenResty

Thank you!Thank you!
Marian HackMan MarinovMarian HackMan Marinov
mm@siteground.commm@siteground.com

4 Sessions

More Related Content

What's hot

Viewers also liked

Similar to 4 Sessions

More from Marian Marinov

Recently uploaded

4 Sessions